ICode9
精准搜索请尝试： 精确搜索
首页
编程语言>
数据库
系统相关
互联网
其他分享
Python
Java
JavaScript
android
PHP
首页 > 其他分享> 文章详细

炫“库”行动-人大金仓有奖征文-递归查询两种写法与性能差异

2021-10-11 10:04:30  阅读：240  来源： 互联网

标签：金仓 rows recursive 征文 pid 有奖 .. test id


【本文正在参与炫“库”行动-人大金仓有奖征文】

活动链接：https://marketing.csdn.net/p/98bd30353e7cb998b6070a89e8b91edb

对于递归查询，KingbaseES 用户可以选择使用connect by ，或者使用 with recursive ，二者在性能上会有差异。下面，我们以例子来看下二者的差别。

一、构造数据

create table test_recursive(id integer,pid integer,name varchar,description text);
insert into test_recursive(id,name,description) select generate_series(1,100000),'a'||generate_series(1,100000),repeat('desc',500);
 
update test_recursive set pid=1 where id between 2 and 10;
update test_recursive set pid=mod(id,9)+2 where id between 11 and 100;
update test_recursive set pid=mod(id,90)+11 where id between 101 and 1000;
update test_recursive set pid=mod(id,900)+101 where id between 1001 and 10000;
update test_recursive set pid=mod(id,9000)+1001 where id between 10001 and 100000;
 
create table test_recursive_random(id integer,pid integer,name varchar,description text);
insert into test_recursive_random select * from test_recursive order by random;
 
create index ind_test_recursive_random_id on test_recursive_random(id);
create index ind_test_recursive_random_pid on test_recursive_random(pid);
vacuum full test_recursive_random;
analyze test_recursive_random;
 
create index ind_test_recursive_id on test_recursive(id);
create index ind_test_recursive_pid on test_recursive(pid);
vacuum full test_recursive;
analyze test_recursive;

本例子构造了5层的数据，有排序与非排序两种数据。

二、使用Connect By

connect by的查询性能：用时 746ms

test=# explain analyze select id,pid,name from test_recursive start with id=1 connect by prior id = pid ;
                                                                        QUERY PLAN                                                                        
-----------------------------------------------------------------------------------------------------------------------------------------------------------
 Recursive Union  (cost=0.29..422.37 rows=101 width=14) (actual time=0.038..728.281 rows=100000 loops=1)
   ->  Index Scan using ind_test_recursive_id on test_recursive  (cost=0.29..8.31 rows=1 width=14) (actual time=0.015..0.017 rows=1 loops=1)
         Index Cond: (id = 1)
   ->  Nested Loop  (cost=0.42..41.30 rows=10 width=14) (actual time=0.002..0.003 rows=1 loops=100000)
         ->  WorkTable Scan on "connect"  (cost=0.00..0.02 rows=1 width=4) (actual time=0.000..0.000 rows=1 loops=100000)
         ->  Index Scan using ind_test_recursive_pid on test_recursive  (cost=0.42..41.18 rows=10 width=14) (actual time=0.002..0.002 rows=1 loops=100000)
               Index Cond: (pid = (PRIOR test_recursive.id))
 Planning Time: 0.185 ms
 Execution Time: 746.102 ms
(9 rows)

三、KingbaseES With Recursive 查询

1、排序数据：用时302ms

explain analyze with recursive tmp1 as (
select id,pid,name from test_recursive where id=1
union all
select a.id,a.pid,a.name from test_recursive a inner join tmp1 b on a.pid=b.id )
select * from tmp1;
                                                                             QUERY PLAN                                                                             
---------------------------------------------------------------------------------------------------------------------------------------------------------------------
 CTE Scan on tmp1  (cost=4013.94..4033.96 rows=1001 width=40) (actual time=0.020..297.856 rows=100000 loops=1)
   CTE tmp1
     ->  Recursive Union  (cost=0.29..4013.94 rows=1001 width=14) (actual time=0.018..257.298 rows=100000 loops=1)
           ->  Index Scan using ind_test_recursive_id on test_recursive  (cost=0.29..8.31 rows=1 width=14) (actual time=0.016..0.018 rows=1 loops=1)
                 Index Cond: (id = 1)
           ->  Nested Loop  (cost=0.42..398.56 rows=100 width=14) (actual time=20.529..38.777 rows=16666 loops=6)
                 ->  WorkTable Scan on tmp1 b  (cost=0.00..0.20 rows=10 width=4) (actual time=0.003..2.150 rows=16667 loops=6)
                 ->  Index Scan using ind_test_recursive_pid on test_recursive a  (cost=0.42..39.74 rows=10 width=14) (actual time=0.001..0.002 rows=1 loops=100000)
                       Index Cond: (pid = b.id)
 Planning Time: 0.207 ms
 Execution Time: 302.244 ms
(11 rows)

2、非排序数据：440ms

test=# explain analyze with recursive tmp1 as (
test(# select id,pid,name from test_recursive_random where id=1
test(# union all
test(# select a.id,a.pid,a.name from test_recursive_random a inner join tmp1 b on a.pid=b.id )
test-# select * from tmp1;
                                                                            QUERY PLAN                                                                            
-------------------------------------------------------------------------------------------------------------------------------------------------------------------
 CTE Scan on tmp1  (cost=4206.87..4226.89 rows=1001 width=40) (actual time=0.020..434.721 rows=100000 loops=1)
   CTE tmp1
     ->  Recursive Union  (cost=0.29..4206.87 rows=1001 width=14) (actual time=0.018..397.456 rows=100000 loops=1)
           ->  Index Scan using ind_test_recursive_random_id on test_recursive_random  (cost=0.29..8.31 rows=1 width=14) (actual time=0.017..0.018 rows=1 loops=1)
                 Index Cond: (id = 1)
           ->  Nested Loop  (cost=4.50..417.85 rows=100 width=14) (actual time=33.080..62.311 rows=16666 loops=6)
                 ->  WorkTable Scan on tmp1 b  (cost=0.00..0.20 rows=10 width=4) (actual time=0.007..2.412 rows=16667 loops=6)
                 ->  Bitmap Heap Scan on test_recursive_random a  (cost=4.50..41.67 rows=10 width=14) (actual time=0.002..0.003 rows=1 loops=100000)
                       Recheck Cond: (pid = b.id)
                       Heap Blocks: exact=99557
                       ->  Bitmap Index Scan on ind_test_recursive_random_pid  (cost=0.00..4.49 rows=10 width=0) (actual time=0.001..0.001 rows=1 loops=100000)
                             Index Cond: (pid = b.id)
 Planning Time: 0.304 ms
 Execution Time: 439.563 ms
(14 rows)

3、使用hash join：260ms

test=# set enable_nestloop=off;
SET
test=# explain analyze with recursive tmp1 as (
test(# select id,pid,name from test_recursive where id=1
test(# union all
test(# select a.id,a.pid,a.name from test_recursive a inner join tmp1 b on a.pid=b.id )
test-# select * from tmp1;
                                                                     QUERY PLAN                                                                     
-----------------------------------------------------------------------------------------------------------------------------------------------------
 CTE Scan on tmp1  (cost=24101.58..24121.60 rows=1001 width=40) (actual time=0.018..255.766 rows=100000 loops=1)
   CTE tmp1
     ->  Recursive Union  (cost=0.29..24101.58 rows=1001 width=14) (actual time=0.016..218.427 rows=100000 loops=1)
           ->  Index Scan using ind_test_recursive_id on test_recursive  (cost=0.29..8.31 rows=1 width=14) (actual time=0.015..0.017 rows=1 loops=1)
                 Index Cond: (id = 1)
           ->  Hash Join  (cost=0.33..2407.32 rows=100 width=14) (actual time=13.828..32.571 rows=16666 loops=6)
                 Hash Cond: (a.pid = b.id)
                 ->  Seq Scan on test_recursive a  (cost=0.00..2031.00 rows=100000 width=14) (actual time=0.005..8.240 rows=100000 loops=6)
                 ->  Hash  (cost=0.20..0.20 rows=10 width=4) (actual time=5.114..5.114 rows=16667 loops=6)
                       Buckets: 131072 (originally 1024)  Batches: 2 (originally 1)  Memory Usage: 3073kB
                       ->  WorkTable Scan on tmp1 b  (cost=0.00..0.20 rows=10 width=4) (actual time=0.004..2.068 rows=16667 loops=6)
 Planning Time: 0.196 ms
 Execution Time: 260.360 ms
(13 rows)

四、执行计划差异分析

  • Connect by 查询执行逻辑：查询是通过 pid = prior id ，也就是将前条记录的 id 作为值，传给 pid 进行索引扫描。逻辑上可以看做是逐个分支逐条记录查询，上个分支查询结束，再进行下个分支扫描。loop = 100000，就是表示针对每条记录，都要访问一次索引。
  • With recursive 查询逻辑：是按层次查询，上层结果都返回后，再执行下层查询。每层可以根据所有ctid进行排序，也就是 Bitmap Index Scan，将所有ctid都返回、排序，再访问表，效率提高。另外，由于是每层数据返回后，再去关联查找下层数据，可以使用hash join，提升访问效率。 rows=16666 loop = 6，表示需要访问6个批次，每次平均 16666 条记录。 

五、Oracle Connect By 查询性能

以下是同样数据量的情况下，Oracle connect by 查询的性能：

SQL> select id,pid,name from test_recursive start with id=1 connect by prior id = pid ;
 
100000 rows selected.
 
Elapsed: 00:00:00.98
 
Execution Plan
----------------------------------------------------------
Plan hash value: 2099392185
 
----------------------------------------------------------------------------------------------------------------
| Id  | Operation                             | Name                   | Rows  | Bytes | Cost (%CPU)| Time     |
----------------------------------------------------------------------------------------------------------------
|   0 | SELECT STATEMENT                      |                        |    12 |   384 |    18  (12)| 00:00:01 |
|*  1 |  CONNECT BY WITH FILTERING            |                        |       |       |            |          |
|   2 |   TABLE ACCESS BY INDEX ROWID BATCHED | TEST_RECURSIVE         |     1 |    32 |     2   (0)| 00:00:01 |
|*  3 |    INDEX RANGE SCAN                   | IND_TEST_RECURSIVE_ID  |     1 |       |     1   (0)| 00:00:01 |
|   4 |   NESTED LOOPS                        |                        |    11 |   495 |    14   (0)| 00:00:01 |
|   5 |    CONNECT BY PUMP                    |                        |       |       |            |          |
|   6 |    TABLE ACCESS BY INDEX ROWID BATCHED| TEST_RECURSIVE         |    11 |   352 |    12   (0)| 00:00:01 |
|*  7 |     INDEX RANGE SCAN                  | IND_TEST_RECURSIVE_PID |    11 |       |     1   (0)| 00:00:01 |
----------------------------------------------------------------------------------------------------------------
 
Predicate Information (identified by operation id):
---------------------------------------------------
 
   1 - access("PID"=PRIOR "ID")
   3 - access("ID"=1)
   7 - access("connect$_by$_pump$_002"."prior id "="PID")
 
Note
-----
   - dynamic statistics used: dynamic sampling (level=2)
   - this is an adaptive plan
 
 
Statistics
----------------------------------------------------------
          0  recursive calls
          0  db block gets
     101983  consistent gets
          0  physical reads
          0  redo size
    2337649  bytes sent via SQL*Net to client
      73769  bytes received via SQL*Net from client
       6668  SQL*Net roundtrips to/from client
          8  sorts (memory)
          0  sorts (disk)
     100000  rows processed

可以看到Oracle 执行耗时大概1秒。 with recursive 针对这种递归查询还是有性能优势的。

【本文正在参与炫“库”行动-人大金仓有奖征文】

活动链接：https://marketing.csdn.net/p/98bd30353e7cb998b6070a89e8b91edb

标签：金仓,rows,recursive,征文,pid,有奖,..,test,id
来源： https://blog.csdn.net/lyu1026/article/details/120696697

本站声明： 1. iCode9 技术分享网（下文简称本站）提供的所有内容，仅供技术学习、探讨和分享；
2. 关于本站的所有留言、评论、转载及引用，纯属内容发起人的个人观点，与本站观点和立场无关；
3. 关于本站的所有言论和文字，纯属内容发起人的个人观点，与本站观点和立场无关；
4. 本站文章均是网友提供，不完全保证技术分享内容的完整性、准确性、时效性、风险性和版权归属；如您发现该文章侵犯了您的权益，可联系我们第一时间进行删除；
5. 本站为非盈利性的个人网站，所有内容不会用来进行牟利，也不会利用任何形式的广告来间接获益，纯粹是为了广大技术爱好者提供技术内容和技术思想的分享性交流网站。

关于我们 | 联系我们 | 留言反馈

专注分享技术，共同学习，共同进步。侵权联系[81616952@qq.com]

Copyright (C)ICode9.com, All Rights Reserved.

ICode9版权所有